Bayesian hierarchical modelling of bridge traffic loading across a road network

Abstract

The prediction of extreme traffic loading is a crucial part of bridge design and assessment. It provides a basis of how much action, and conversely the strength required for a bridge in its lifetime. However, there remain large uncertainties in the predictions of extreme loading due to the complex nature of the underlying traffic and its load effects. So far, efforts to model extreme load effects have been done primarily using standard extreme value methods, such as the generalized extreme value distribution and generalized Pareto distribution. However, these efforts provide techniques for fitting data from a single bridge only, requiring extrapolations for predictions to other bridge spans, which carries large uncertainties. Single fit methods also fail to take advantage of information contained in other spans that could be used to reduce estimation uncertainties ﾖ the ﾑshrinkageﾒ effect. In this paper, a modern Bayesian hierarchical model is developed using the generalized extreme value distribution, covering intermediate spans where data is not available at the time of fitting. First, simple Bayesian model was explained and used with a simple and considered realistic traffic model. The simple model is then expanded to a Bayesian hierarchical model to simultaneously fit a range of spans. The final model shows accurate predictions for intermediate spans not used for the fitting process and provides reduced uncertainties compared to the single-span fits. This work provides a basis for estimating load effects across an entire road network at once; something not previously feasible.